Spaces:
Sleeping
Sleeping
Update ViT_DeiT/baselines/ViT/ViT_explanation_generator.py
Browse files
ViT_DeiT/baselines/ViT/ViT_explanation_generator.py
CHANGED
@@ -1,28 +1,46 @@
|
|
1 |
import argparse
|
2 |
-
|
3 |
import numpy as np
|
|
|
4 |
from numpy import *
|
5 |
|
|
|
6 |
# compute rollout between attention layers
|
7 |
def compute_rollout_attention(all_layer_matrices, start_layer=0):
|
8 |
# adding residual consideration- code adapted from https://github.com/samiraabnar/attention_flow
|
9 |
num_tokens = all_layer_matrices[0].shape[1]
|
10 |
batch_size = all_layer_matrices[0].shape[0]
|
11 |
-
eye =
|
12 |
-
|
13 |
-
|
14 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
joint_attention = matrices_aug[start_layer]
|
16 |
-
for i in range(start_layer+1, len(matrices_aug)):
|
17 |
joint_attention = matrices_aug[i].bmm(joint_attention)
|
18 |
return joint_attention
|
19 |
|
|
|
20 |
class LRP:
|
21 |
def __init__(self, model):
|
22 |
self.model = model
|
23 |
self.model.eval()
|
24 |
|
25 |
-
def generate_LRP(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
26 |
output = self.model(input)
|
27 |
kwargs = {"alpha": 1}
|
28 |
if index == None:
|
@@ -32,14 +50,18 @@ class LRP:
|
|
32 |
one_hot[0, index] = 1
|
33 |
one_hot_vector = one_hot
|
34 |
one_hot = torch.from_numpy(one_hot).requires_grad_(True)
|
35 |
-
one_hot = torch.sum(one_hot
|
36 |
|
37 |
self.model.zero_grad()
|
38 |
one_hot.backward(retain_graph=True)
|
39 |
|
40 |
-
return self.model.relprop(
|
41 |
-
|
42 |
-
|
|
|
|
|
|
|
|
|
43 |
|
44 |
|
45 |
class Baselines:
|
@@ -48,14 +70,14 @@ class Baselines:
|
|
48 |
self.model.eval()
|
49 |
|
50 |
def generate_cam_attn(self, input, index=None):
|
51 |
-
output = self.model(input
|
52 |
if index == None:
|
53 |
index = np.argmax(output.cpu().data.numpy())
|
54 |
|
55 |
one_hot = np.zeros((1, output.size()[-1]), dtype=np.float32)
|
56 |
one_hot[0][index] = 1
|
57 |
one_hot = torch.from_numpy(one_hot).requires_grad_(True)
|
58 |
-
one_hot = torch.sum(one_hot
|
59 |
|
60 |
self.model.zero_grad()
|
61 |
one_hot.backward(retain_graph=True)
|
@@ -79,5 +101,7 @@ class Baselines:
|
|
79 |
attn_heads = blk.attn.get_attention_map()
|
80 |
avg_heads = (attn_heads.sum(dim=1) / attn_heads.shape[1]).detach()
|
81 |
all_layer_attentions.append(avg_heads)
|
82 |
-
rollout = compute_rollout_attention(
|
83 |
-
|
|
|
|
|
|
1 |
import argparse
|
2 |
+
|
3 |
import numpy as np
|
4 |
+
import torch
|
5 |
from numpy import *
|
6 |
|
7 |
+
|
8 |
# compute rollout between attention layers
|
9 |
def compute_rollout_attention(all_layer_matrices, start_layer=0):
|
10 |
# adding residual consideration- code adapted from https://github.com/samiraabnar/attention_flow
|
11 |
num_tokens = all_layer_matrices[0].shape[1]
|
12 |
batch_size = all_layer_matrices[0].shape[0]
|
13 |
+
eye = (
|
14 |
+
torch.eye(num_tokens)
|
15 |
+
.expand(batch_size, num_tokens, num_tokens)
|
16 |
+
.to(all_layer_matrices[0].device)
|
17 |
+
)
|
18 |
+
all_layer_matrices = [
|
19 |
+
all_layer_matrices[i] + eye for i in range(len(all_layer_matrices))
|
20 |
+
]
|
21 |
+
matrices_aug = [
|
22 |
+
all_layer_matrices[i] / all_layer_matrices[i].sum(dim=-1, keepdim=True)
|
23 |
+
for i in range(len(all_layer_matrices))
|
24 |
+
]
|
25 |
joint_attention = matrices_aug[start_layer]
|
26 |
+
for i in range(start_layer + 1, len(matrices_aug)):
|
27 |
joint_attention = matrices_aug[i].bmm(joint_attention)
|
28 |
return joint_attention
|
29 |
|
30 |
+
|
31 |
class LRP:
|
32 |
def __init__(self, model):
|
33 |
self.model = model
|
34 |
self.model.eval()
|
35 |
|
36 |
+
def generate_LRP(
|
37 |
+
self,
|
38 |
+
input,
|
39 |
+
index=None,
|
40 |
+
method="transformer_attribution",
|
41 |
+
is_ablation=False,
|
42 |
+
start_layer=0,
|
43 |
+
):
|
44 |
output = self.model(input)
|
45 |
kwargs = {"alpha": 1}
|
46 |
if index == None:
|
|
|
50 |
one_hot[0, index] = 1
|
51 |
one_hot_vector = one_hot
|
52 |
one_hot = torch.from_numpy(one_hot).requires_grad_(True)
|
53 |
+
one_hot = torch.sum(one_hot * output)
|
54 |
|
55 |
self.model.zero_grad()
|
56 |
one_hot.backward(retain_graph=True)
|
57 |
|
58 |
+
return self.model.relprop(
|
59 |
+
torch.tensor(one_hot_vector).to(input.device),
|
60 |
+
method=method,
|
61 |
+
is_ablation=is_ablation,
|
62 |
+
start_layer=start_layer,
|
63 |
+
**kwargs
|
64 |
+
)
|
65 |
|
66 |
|
67 |
class Baselines:
|
|
|
70 |
self.model.eval()
|
71 |
|
72 |
def generate_cam_attn(self, input, index=None):
|
73 |
+
output = self.model(input, register_hook=True)
|
74 |
if index == None:
|
75 |
index = np.argmax(output.cpu().data.numpy())
|
76 |
|
77 |
one_hot = np.zeros((1, output.size()[-1]), dtype=np.float32)
|
78 |
one_hot[0][index] = 1
|
79 |
one_hot = torch.from_numpy(one_hot).requires_grad_(True)
|
80 |
+
one_hot = torch.sum(one_hot * output)
|
81 |
|
82 |
self.model.zero_grad()
|
83 |
one_hot.backward(retain_graph=True)
|
|
|
101 |
attn_heads = blk.attn.get_attention_map()
|
102 |
avg_heads = (attn_heads.sum(dim=1) / attn_heads.shape[1]).detach()
|
103 |
all_layer_attentions.append(avg_heads)
|
104 |
+
rollout = compute_rollout_attention(
|
105 |
+
all_layer_attentions, start_layer=start_layer
|
106 |
+
)
|
107 |
+
return rollout[:, 0, 1:]
|